18047796. SELF-SUPERVISED THREE-DIMENSIONAL LOCATION PREDICTION USING MACHINE LEARNING MODELS simplified abstract (QUALCOMM Incorporated)
SELF-SUPERVISED THREE-DIMENSIONAL LOCATION PREDICTION USING MACHINE LEARNING MODELS
Organization Name
Inventor(s)
Hanno Ackermann of Hilversum (NL)
Ilia Karmanov of Amsterdm (NL)
Farhad Ghazvinian Zanjani of Almere (NL)
Daniel Hendricus Franciscus Dijkman of Haarlem (NL)
Fatih Murat Porikli of San Diego CA (US)
SELF-SUPERVISED THREE-DIMENSIONAL LOCATION PREDICTION USING MACHINE LEARNING MODELS - A simplified explanation of the abstract
This abstract first appeared for US patent application 18047796 titled 'SELF-SUPERVISED THREE-DIMENSIONAL LOCATION PREDICTION USING MACHINE LEARNING MODELS
Simplified Explanation
The present disclosure describes a method for training a machine learning model to predict the location of a device in a spatial environment.
- The method involves training a generator model to map scene data to points in three-dimensional space.
- Critic models are trained to provide feedback to the generator model, pushing the points in three-dimensional space towards specific planes.
- The generator model is then deployed for predicting the location of a device in a spatial environment.
Potential Applications
This technology has potential applications in various fields, including:
- Robotics: The ability to predict the location of a device in a spatial environment can be valuable for autonomous robots navigating in complex environments.
- Augmented Reality: Predicting device location accurately can enhance the user experience in augmented reality applications by aligning virtual objects with the real world.
- Indoor Navigation: This technology can assist in indoor navigation systems by accurately determining the location of a device within a building or facility.
Problems Solved
The self-supervised training method described in the patent application addresses the following problems:
- Device Localization: By training a machine learning model to predict device location in a spatial environment, this technology solves the problem of accurately determining the position of a device.
- Spatial Understanding: The method helps the machine learning model understand the spatial layout of an environment, including multiple planes, enabling better predictions and decision-making.
Benefits
The use of self-supervised training for device localization offers several benefits:
- Improved Accuracy: By training the model to map scene data to points in three-dimensional space, the accuracy of device location prediction can be significantly improved.
- Versatility: The method can be applied to various spatial environments, including those with multiple discrete planes.
- Real-time Prediction: Once deployed, the trained model can provide real-time predictions of device location, enabling immediate decision-making and action.
Original Abstract Submitted
Certain aspects of the present disclosure provide techniques method for self-supervised training of a machine learning model to predict the location of a device in a spatial environment, such as a spatial environment including multiple discrete planes. An example method generally includes receiving an input data set of scene data. A generator model is trained to map scene data in the input data set to points in three-dimensional space. One or more critic models are trained to backpropagate a gradient to the generator model to push the points in the three-dimensional space to one of a plurality of planes in the three-dimensional space. At least the generator is deployed.
